Full Automatic Relative Orientation of Unconventional Image Pairs

Relative orientation is a very important step during photogrammetry. Traditional relative orientation is realized by adjusting the relative elements value, and eliminating vertical parallax of outstanding points on six standard point positions, or by analytical calculation after relative orientation with image point coordinates which are outstanding points on six standard point positions according to manual observation. With the development of digital photogrammetric technology, full automatic relative orientation can be implemented automatically by computer. There are mainly two steps: the first step is relative orientation observation acquisition by feature points extraction and matching; the other step is to determine direct position relationship using relative orientation elements which calculated with relative observation. However, feature point extraction and matching are the key of full automatic relative orientation. This paper discusses that making use of scale invariant feature transform (SIFT) operator to determine the initial position relationship between images under the circumstance of different photography. Furthermore, it can guide high precise image matching operator to least squares image matching. It proves that full automatic relative orientation can be realized in this way after enough corresponding observations.